Snapshots in Ceph, deleting an "active" snapshot

Community Support Storage
1

Hi all,

there still seems to be something fundamentally wrong in how OpenNebula handles snapshots in Ceph: I would like to use e.g. daily snapshots as a “lightweight backup”, to be able to revert to them later, should something go wrong with a long-running VM. And keep three or four latest snapshots.

$ onevm disk-snapshot-create 7317 0 testsnap
$ onevm disk-snapshot-delete 7317 0 testsnap

OK, snapshots can be created and deleted. Now let’s test reverting to the previous state:

$ onevm disk-snapshot-create 7317 0 testsnap2
$ ssh root@vm7317 'touch /root/testfile'
$ onevm poweroff 7317
$ onevm disk-snapshot-revert 7317 0 testsnap2
$ onevm resume 7317
$ ssh root@vm7317 'ls -l /root/testfile'
ls: cannot access '/root/testfile': No such file or directory
$ onevm disk-snapshot-delete 7317 0 testsnap2
[one.vm.disksnapshotdelete] Cannot delete the active snapshot

After reverting to a snapshot, it somehow becomes “active”, and cannot be deleted anymore. Nevermind, let’s create another “backup” snapshot:

$ onevm disk-snapshot-create 7317 0 testsnap3
$ onevm poweroff 7317
$ onevm disk-snapshot-revert testsnap3
$ onevm resume 7317
$ onevm disk-snapshot-delete 7317 0 testsnap2
[one.vm.disksnapshotdelete] Cannot delete snapshot with children

It seems that OpenNebula does not use Ceph snapshots at all, but uses RBD image clones instead:

$ rbd snap ls one/one-1920
... no snapshots at all!
$ rbd ls one|grep 1920
one-1920
one-1920-0
one-1920-1
$ rbd snap ls one/one-1920-0
SNAPID  NAME  SIZE   PROTECTED  TIMESTAMP               
  2457  0     4 GiB  yes        Fri Jun  2 10:31:47 2023
$ rbd snap ls one/one-1920-1
SNAPID  NAME  SIZE   PROTECTED  TIMESTAMP               
  2458  1     4 GiB  yes        Fri Jun  2 10:39:01 2023

Why is it done this way? I would expect ONe snapshots to use Ceph RBD snapshots directly in a flat structure. But nevermind the implementation, I would definitely not expect reverting to the previous snapshot locking that snapshot forever.

Do you consider this a bug? Other virtualization platforms allow to snapshot a VM disk easily, e.g. before the upgrade, revert to it if something goes wrong, and delete a snapshot whenever it is not needed anymore.

How can I delete testsnap2 and testsnap3 above, without disturbing the state of the running VM #7317 ?

Thanks!

-Yenya